Thermal Modeling, Analysis, and Management in VLSI Circuits: Principles and Methods
read more
Citations
Nanoscale thermal transport. II. 2003–2012
A critical review of traditional and emerging techniques and fluids for electronics cooling
Underdesigned and Opportunistic Computing in Presence of Hardware Variability
Experimental investigation on paraffin wax integrated with copper foam based heat sinks for electronic components thermal cooling
Predictive dynamic thermal and power management for heterogeneous mobile platforms
References
I and i
From the authors
Wattch: a framework for architectural-level power analysis and optimizations
Fundamentals of Modern VLSI Devices
Electromigration—A brief survey and some recent results
Related Papers (5)
Frequently Asked Questions (20)
Q2. How much delay degradation is achieved for each 20°C temperature increase along the interconnect line?
Assuming a fixed driver resistance, which simplifies the delay expression, the authors show that about 5% delay degradation is achieved for each 20°C temperature increase in the constant temperature along the interconnect line.
Q3. What are the possible response mechanisms to the thermal impact of a chip?
Possible response mechanisms include micro-architectural adaptations (e.g., clock throttling, register file resizing, limiting the issue width of a processor, and computation migration to auxiliary hardware), and/or on-the-fly performance adjustment via dynamic power management (DPM), dynamic voltage scaling (DVS), clock/power gating.
Q4. Why are voltage drop effects becoming increasingly significant?
Voltage drop effects are becoming increasingly significant, because the resistivity of the power and ground tracks rises as a function of decreasing feature sizes (track widths) and increasing chip temperatures.
Q5. What are the advantages of using dummy vias in the higher metal layers?
Dummy vias in the higher metal layers may be used to reduce temperatures on interconnect without impacting their electrical resistance and capacitance.
Q6. Why is thermal management accomplished when it is incorporated at the beginning of the design cycle?
Because heat conduction is playing a bigger role while heat convection is playing a lesser role in removing heat, thermal management is best accomplished when it is incorporated starting at the beginning of the design cycle.
Q7. What is the effect of thermal gradients on the interconnects?
The existence of high thermal gradients on the substrate creates non-uniform temperature profiles along the length of the global interconnect lines, which are located above the substrate.
Q8. How does the temperature of metals increase beyond the 45 nm node?
metal temperatures increase significantly (i.e., by hundreds of Celsius degrees) beyond the 45-nm node owing to the combined effects of increasing metal resistivity, increasing current density, increasing number of global metal levels, and decreasing ILD thermal conductivity.
Q9. How can designers make designs more robust?
Designers can also make designs more robust by limiting the maximum power draw that is sustained over a period of say tens of micro-seconds.
Q10. Why is the heat diffusion length larger for the higher level metal layers?
Assuming a constant current density in all metal layers of a signal net, it is found that the heat diffusion length is larger for the higher level metal layers due to their higher underlying insulator thickness.
Q11. What is the common cause of voltage drops on the power rail?
In general, the voltage drops on the power rail can be in the form of a self-induced IR drop from the external power pin to the power terminal of a logic block due to the current that is drawn by the logic block itself.
Q12. Why does the VT drop reduce the noise margins of logic cells?
This is in turn because variations in VT or Leff of transistors and voltage drops on power supply lines reduce the noise margins of logic cells, leaving less room to accommodate temperature effects on parasitic RC values and circuit performance.
Q13. What is the impact of the non-uniform resistance profile of the global interconnects?
The non-uniform resistance profile of the global interconnects will in turn strongly impact many aspects of interconnect performance modeling and optimization.
Q14. What is the effect of thermal runaway on the chip?
Even when thermal runaway does not occur, the operating temperature of the chip can become larger than the designed value, which will either increase the package cost or degrade the performance as well as the long-term reliability of the chip [9].
Q15. How can the effect of voltage drop be minimized?
These effects can be minimized by increasing the width of power tracks (which reduces the power track resistances) and/or by increasing the spacing between logic blocks (which reduces power density and hence reduces chip temperature).
Q16. What is the heat diffusion equation for a metal interconnect?
The net heat energy generation per unit volume is:( ) ( ) ( , ) gen lossmP x P x g x Twt x− =Δ (11)Since the length of global interconnects can be assumed to be much larger than its thickness and width, the thermal gradients along thickness and width can be ignored, i.e., a 1-D formula can be found to represent the heat diffusion:
Q17. How can one find the thermal profile of interconnection lines?
by solving eqn. (12) subject to boundary condition (i.e., given temperatures at the two ends of the line), the thermal profile of interconnection lines can be found.
Q18. What is the simplest way to model the die temperature?
As will be shown in Section 4, for a given package, the die temperature can be modeled as a linear function of the total power dissipation of the circuit.
Q19. How does the graph show the increase in leakage power of a die?
Figure 1 illustrates the significant increase in leakage power of a 15mm die fabricated in a 100nm technology with a supply voltage of 0.7V as a function of substrate temperature.
Q20. Why were corner-based techniques used as relatively fast techniques?
Up until recently, corner-based timing and signal integrity analysis techniques were used as relatively fast techniques to address the concerns related to various sources of variation in VLSI circuits.